Methods, apparatus, and articles of manufacture for interactive image segmentation

Abstract

Methods, apparatus, systems, and articles of manufacture are disclosed for interactive image segmentation. An example apparatus includes an inception controller to execute an inception sublayer of a convolutional neural network (CNN) including two or more inception-atrous-collation (IAC) layers, the inception sublayer including two or more convolutions including respective kernels of varying sizes to generate multi-scale inception features, the inception sublayer to receive one or more context features indicative of user input; an atrous controller to execute an atrous sublayer of the CNN, the atrous sublayer including two or more atrous convolutions including respective kernels of varying sizes to generate multi-scale atrous features; and a collation controller to execute a collation sublayer of the CNN to collate the multi-scale inception features, the multi-scale atrous features, and eidetic memory features.

What is claimed is: 
     
       1. An apparatus for interactive image segmentation, the apparatus comprising:
 an inception controller to execute an inception sublayer of a convolutional neural network (CNN) including two or more inception-atrous-collation (IAC) layers, the inception sublayer including two or more convolutions including respective kernels of varying sizes to generate multi-scale inception features, the inception sublayer to receive one or more context features indicative of user input; 
 an atrous controller to execute an atrous sublayer of the CNN, the atrous sublayer including two or more atrous convolutions including respective kernels of varying sizes to generate multi-scale atrous features; and 
 a collation controller to execute a collation sublayer of the CNN to collate the multi-scale inception features, the multi-scale atrous features, and eidetic memory features. 
 
     
     
       2. The apparatus of  claim 1 , wherein the inception controller is to bypass the inception sublayer in response to determining that bypassing the inception sublayer would be advantageous to the CNN. 
     
     
       3. The apparatus of  claim 2 , wherein the inception controller is to bypass the inception sublayer to preserve data in one or more input feature matrices to the inception sublayer. 
     
     
       4. The apparatus of  claim 1 , wherein the atrous controller is to bypass the atrous sublayer in response to determining that bypassing the atrous sublayer would be advantageous to the CNN. 
     
     
       5. The apparatus of  claim 4 , wherein the atrous controller is to bypass the atrous sublayer to preserve data in the inception features. 
     
     
       6. The apparatus of  claim 1 , wherein the collation controller is to bypass a subsequent IAC layer in response to determining that bypassing the subsequent IAC layer would be advantageous to the CNN. 
     
     
       7. The apparatus of  claim 1 , wherein the user input corresponds to one or more selections in an image to be processed, the one or more selections identifying one or more pixels within a threshold distance of one or more respective selection epicenters and a corresponding respective gradient specifying a likelihood that the one or more pixels are within the threshold distance. 
     
     
       8. A non-transitory computer-readable medium comprising instructions which, when executed, cause at least one processor to at least:
 implement an inception sublayer of a convolutional neural network (CNN) including two or more inception-atrous-collation (IAC) layers, the inception sublayer including two or more convolutions including respective kernels of varying sizes to generate multi-scale inception features, the inception sublayer to receive one or more context features indicative of user input; 
 implement an atrous sublayer of the CNN, the atrous sublayer including two or more atrous convolutions including respective kernels of varying sizes to generate multi-scale atrous features; and 
 implement a collation sublayer of the CNN to collate the multi-scale inception features, the multi-scale atrous features, and eidetic memory features. 
 
     
     
       9. The non-transitory computer-readable medium of  claim 8 , wherein the instructions, when executed, cause the at least one processor to bypass the inception sublayer in response to determining that bypassing the inception sublayer would be advantageous to the CNN. 
     
     
       10. The non-transitory computer-readable medium of  claim 9 , wherein the instructions, when executed, cause the at least one processor to bypass the inception sublayer to preserve data in one or more input feature matrices to the inception sublayer. 
     
     
       11. The non-transitory computer-readable medium of  claim 8 , wherein the instructions, when executed, cause the at least one processor to bypass the atrous sublayer in response to determining that bypassing the atrous sublayer would be advantageous to the CNN. 
     
     
       12. The non-transitory computer-readable medium of  claim 11 , wherein the instructions, when executed, cause the at least one processor to bypass the atrous sublayer to preserve data in the inception features. 
     
     
       13. The non-transitory computer-readable medium of  claim 8 , wherein the instructions, when executed, cause the at least one processor to bypass a subsequent IAC layer in response to determining that bypassing the subsequent IAC layer would be advantageous to the CNN. 
     
     
       14. The non-transitory computer-readable medium of  claim 8 , wherein the user input corresponds to one or more selections in an image to be processed, the one or more selections identifying one or more pixels within a threshold distance of one or more respective selection epicenters and a corresponding respective gradient specifying a likelihood that the one or more pixels are within the threshold distance. 
     
     
       15. An apparatus for interactive image segmentation, the apparatus comprising:
 memory; and 
 at least one processor to execute machine readable instructions to cause the at least one processor to:
 implement an inception sublayer of a convolutional neural network (CNN) including two or more inception-atrous-collation (IAC) layers, the inception sublayer including two or more convolutions including respective kernels of varying sizes to generate multi-scale inception features, the inception sublayer to receive one or more context features indicative of user input; 
 implement an atrous sublayer of the CNN, the atrous sublayer including two or more atrous convolutions including respective kernels of varying sizes to generate multi-scale atrous features; and 
 implement a collation sublayer of the CNN to collate the multi-scale inception features, the multi-scale atrous features, and eidetic memory features. 
 
 
     
     
       16. The apparatus of  claim 15 , wherein the at least one processor is to bypass the inception sublayer in response to determining that bypassing the inception sublayer would be advantageous to the CNN. 
     
     
       17. The apparatus of  claim 16 , wherein the at least one processor is to bypass the inception sublayer to preserve data in one or more input feature matrices to the inception sublayer. 
     
     
       18. The apparatus of  claim 15 , wherein the at least one processor is to bypass the atrous sublayer in response to determining that bypassing the atrous sublayer would be advantageous to the CNN. 
     
     
       19. The apparatus of  claim 18 , wherein the at least one processor is to bypass the atrous sublayer to preserve data in the inception features. 
     
     
       20. The apparatus of  claim 15 , wherein the at least one processor is to bypass a subsequent IAC layer in response to determining that bypassing the subsequent IAC layer would be advantageous to the CNN. 
     
     
       21. The apparatus of  claim 15 , wherein the user input corresponds to one or more selections in an image to be processed, the one or more selections identifying one or more pixels within a threshold distance of one or more respective selection epicenters and a corresponding respective gradient specifying a likelihood that the one or more pixels are within the threshold distance. 
     
     
       22. A method for interactive image segmentation, the method comprising:
 executing an inception sublayer of a convolutional neural network (CNN) including two or more inception-atrous-collation (IAC) layers, the inception sublayer including two or more convolutions including respective kernels of varying sizes to generate multi-scale inception features, the inception sublayer to receive one or more context features indicative of user input; 
 executing an atrous sublayer of the CNN, the atrous sublayer including two or more atrous convolutions including respective kernels of varying sizes to generate multi-scale atrous features; and 
 executing a collation sublayer of the CNN to collate the multi-scale inception features, the multi-scale atrous features, and eidetic memory features. 
 
     
     
       23. The method of  claim 22 , further including bypassing the inception sublayer in response to determining that bypassing the inception sublayer would be advantageous to the CNN. 
     
     
       24. The method of  claim 23 , further including bypassing the inception sublayer to preserve data in one or more input feature matrices to the inception sublayer. 
     
     
       25. The method of  claim 22 , further including bypassing the atrous sublayer in response to determining that bypassing the atrous sublayer would be advantageous to the CNN.